There's (still) plenty of room at the bottom

There's (still) plenty of room at the bottom

Available online at www.sciencedirect.com ScienceDirect There’s (still) plenty of room at the bottom Noah Olsman1 and Lea Goentoro2 Motifs, circuits,...

1MB Sizes 0 Downloads 46 Views

Available online at www.sciencedirect.com

ScienceDirect There’s (still) plenty of room at the bottom Noah Olsman1 and Lea Goentoro2 Motifs, circuits, and networks are core conceptual elements in modern systems and synthetic biology. While there are still undoubtedly more fascinating computations to discover at network level, there are also rich computations that we are only beginning to uncover within the diverse molecules that constitute the networks. Here we explore some work, both new and old, that showcases the incredible computational capacity of seemingly simple molecular mechanisms. A more sophisticated understanding of computations at the molecular level will inspire the development of a more nuanced toolbox for future biological engineering.

Addresses 1 Department of Computing and Mathematical Sciences, California Institute of Technology, United States 2 Division of Biology and Biological Engineering, California Institute of Technology, United States Corresponding authors: Olsman, Noah ([email protected]), Goentoro, Lea ([email protected])

great deal of information must be processed using DNA, RNA, and proteins. When studying information processing in biology, we often focus on the computational capacity of circuits and motifs of a few components [2] and networks of tens to hundreds of elements [3–5]. While there is certainly still much to be explored at the level of circuits and networks, this perspective often coarse-grains the finer molecular details of these systems. Our viewpoint here will zoom in on these often-overlooked molecular details, and highlight several case studies where molecules perform an impressive range of computation. For instance, while a membrane receptor can simply be viewed as serving to transduce signal from outside the cell, a closer examination reveals that it can be an adaptive component in signaling, performing a wide range of nonlinear computation. Similarly, mRNA is a template for protein translation, but a closer look reveals it to be a powerful regulatory hub, integrating a variety of chemical and environmental stimuli.

Current Opinion in Structural Biology 2018, 54:72–79

An enzymatic proofreader

This review comes from a themed issue on Analytical biotechnology

We begin with a seminal paper by John Hopfield, published in 1973 [6], that asked a simple question: how is it that the synthesis of biological molecules has such a low rate of error? Hopfield considered a simple model of protein translation. Protein translation proceeds by an mRNA template being processed by a ribosome, which recruits tRNA that ultimately attach amino acids to the nascent chain. A key step is when a ligase attaches an amino acid to a transfer RNA (tRNA). If the ligase attaches the wrong amino acid then, even if the correct tRNA recruited, there will be an error in protein translation.

Edited by Hiroshi Shimizu and Fumio Matsuda

https://doi.org/10.1016/j.copbio.2018.01.029 0958-1669/ã 2018 Elsevier Ltd. All rights reserved.

Exploring the molecular toolbox Almost 60 years ago, Richard Feynman gave a nowfamous lecture titled ‘There’s Plenty of Room at the Bottom’ [1], where he argued that we had only just begun to understand the extent to which the physical world can be manipulated at the molecular scale. He expressed wonderment at processes like photosynthesis and the translation of proteins. At the time, relatively little was known about the structure, function, and organization of the molecules that underly these phenomena, but it was clear from their design that evolution is a resourceful engineer (Figure 1). In the last few decades, Feynman’s vision has largely become a reality. We now understand, in a much deeper way, that single cells have an astonishing capacity to sense and make decisions about their environment. From quorum sensing in bacteria to embryonic development in animals, a Current Opinion in Biotechnology 2018, 54:72–79

Hopfield first analyzed the expected error rate due to the ligation of incorrect amino acids to tRNA. Take, for example, equal concentrations of the amino acids isoleucine and valine. The binding affinity of the isoleucine ligase for the former is about 100 that of the latter, implying an approximate error rate of 102. With more types of amino acids, one would expect the error rate to only get worse. These results are in stark contrast to the real error rate in protein translation, which is closer to 104 [7]. The key to resolving this discrepancy is the phenomenon that Hopfield referred to as kinetic proofreading. To motivate this idea, imagine a point during the assembly of a protein where amino acid is required, however there is an equal amount of amino acid available, described by the reactions in Figure 2a. Let us assume that the www.sciencedirect.com

There’s (still) plenty of room at the bottom Olsman and Goentoro 73

Figure 1

Molecules

Mechanisms

Functions Logarithmic Sensor

f(x) = log(x) 42˚C

Translation

RNA Thermometer 30˚C

Temperature Current Opinion in Biotechnology

Molecules, mechanisms, and functions. A primary goal of systems and synthetic biology is to gain a deep understanding of the connection between the molecules that make up biological processes and their associated functions. While we often focus on large biomolecular networks, we can often gain insight from studying the quantitative functional properties of individual molecular mechanisms. In the row, we see a protein with multiple subunits, the mechanism of conformational switching, and the function of responding to signal in a logarithmic fashion. On the bottom row, we see an RNA with a particular secondary structure, the mechanism of temperature regulating the availability of the ribosome binding site, and the function of protein translation being a function of temperature.

ligase has a relative dissociation rate . This would yield the 102 error rate mentioned earlier. Now imagine adding a second step that we will refer to as thermodynamic proofreading. Instead of directly adding an amino acid, we will instead have an intermediary stage where some form of post-translational modification takes place. Ideally this intermediate stage would allow for additional ligand specificity, giving the ligase an extra opportunity to ameliorate erroneous binding (Figure 2b). Unfortunately, at equilibrium this cannot be the case. Hopfield showed explicitly that due to the constraints of thermodynamic equilibrium, while the second step does preferentially reverse off target binding, the nature of the comfirst step implies that there will be far more complexes. These rates must balance in plexes than such a way that the rate of erroneous binding is never less than the original 102. Since the impediment comes from thermodynamic constraints, it makes sense to ask if we can somehow sidestep them. One way to do this is to pump energy into the system, by some ATP-consuming enzymatic process, which means we are in the regime of chemical kinetics (Figure 2c). This frees us of thermodynamic equilibrium constraints and makes it possible to reduce error rates by a factor of the previous rate squared, or 104. The addition of more successive energy-intensive steps would continue to drop the error rate. Over the last 40 years kinetic proofreading has emerged as a pervasive mechanism in various biological pathways, from T-cell receptor signal www.sciencedirect.com

transduction [8] to chromatin remodeling [9,10]. On the theoretical side, recent work has vastly expanded our understanding of how kinetic proofreading works from the perspective of non-equilibrium statistical mechanics [11]. While there is no free lunch in life, sometimes it is worth paying a bit extra to get a better meal.

An RNA thermometer Next we shift focus from preventing errors to preventing catastrophes. Every known organism on Earth has some mechanism for responding to a sudden increase in temperature, referred to as heat shock response (HSR) [12]. One reason heat shock is dangerous is that it can cause proteins to become misfolded, which can be lethal for a cell. To avoid this, the HSR system senses temperature change, and synthesizes chaperone proteins whose job is it is to refold proteins [13]. How does a cell sense when it needs to make chaperones? One way is to sense misfolded proteins (i.e. feedback response), the other is to sense temperature changes so that chaperone synthesis can begin before misfolded proteins have already accumulated (i.e. feedforward response). While both sensing mechanisms exist [14,15], we focus here on the latter feedforward response that is mediated by messenger RNA. Since heat shock response needs to be fast, it is reasonable to posit that any temperature sensing mechanism must be directly linked to the synthesis of chaperones. To solve this, Escherichia coli have evolved an ingenious Current Opinion in Biotechnology 2018, 54:72–79

74 Analytical biotechnology

Figure 2

Target Sequence:

Error Rate

Chemical Reactions

Assembly Mechanism (a) No Proofreading:

k’ k

W

k’ k

W

10-2

(b) t

t’

k’ k

m’ m

W

k’ k

m’ m

W

Thermodynamic Proofreading:

10-2 t

t’

t

t’

(c)

k’ k

Kinetic Proofreading: k’ k

ATP

W

m,m’

10-4

ATP

W

m,m’ t

t’

Current Opinion in Biotechnology

Proofreading through enzyme kinetics. (a) A naive approach to protein translation. Amino acids bind to corresponding ligases with various specific dissociation constants K D ¼ kk0 . Differing values of KD provide the only source of specificity in terms of amino acid ligation (which occurs at the non-specific rate W). (b) A slightly more sophisticated approach to proofreading, where there is now a non-specific step mediated by rates m and m0 which modifies the amino acids and a specific step mediated by t and t0 . If all reactions are purely thermodynamic, the error rate for the setting is the same as in (a). (c) Here we see Hopfield’s kinetic proofreading. The only difference here is that the non-specific modification at rates m and m0 uses external energy. This kinetic process will reduce the error rate below the thermodynamic boundary that constrained the first two processes.

system. Their HSR is mediated by a single sigma factor s 32, which regulates the production of dozens of HSR-related proteins [16]. In order to make s 32 expression temperature dependent, the s 32 mRNA has a hairpin structure which blocks the ribosome binding site at nominal temperatures [17,18]. At higher temperatures, the hairpin becomes unstable and consequently allows for initiation of translation. Moreover, because of the Current Opinion in Biotechnology 2018, 54:72–79

thermodynamic nature of this transition, the increase in translation exhibits a graded response over a range of temperatures. Therefore, the s 32 mRNA acts as a thermometer, and is an example of the many ways biology regulates mRNA secondary structure to quantitatively control protein levels. An analogous mechanism is the riboswitch, where www.sciencedirect.com

There’s (still) plenty of room at the bottom Olsman and Goentoro 75

Figure 3

Unbound

(a) KA

KA

Active

Bound

e ε0 KI

KI

Inactive

(b)

Fold-Change Detector

KA

KA

KI

Signal

Logarithmic Sensor

c

a(c,ε0)

ε0

KI Negative Feedback

Activity

(c)

Basal Activity 1

1

0.5

25 25

50

0

100

25

50

0.5 0

100

1

50 25

Tuning of ε0

50

100 0.5

100

0

Signal Level

25

50

100

25

50

100

1

Signal Change:

0.5 0

Tuning of ε0

50 0.5

1

Activity

1

Signal Change:

0

Adaptation

Fold Change in Signal

25

50

0.5 0

100

Signal Level

Signal Level Current Opinion in Biotechnology

Logarithmic computation through allostery. (a) A schematic diagram of the possible states for an allosteric protein. This protein has four subunits, each of which can either be bound or unbound. The active and inactive states have different KD values, and there exists a mechanism for the states to switch back and forth on a concerted way at a rate determined by the allosteric parameter ee0. (b) The general architecture of a circuit that can use an allosteric protein to implement fold-change detection. (c) The top row shows an allosteric protein at a basal level of activation. A signal change then occurs from 25 to 50, increasing the activity. In the final step, the activity of the protein is allosteric regulated such that its activation returns to basal. Because the allosteric regulation effectively moves the response curve in log space, we see exactly the same dynamics in the bottom row when the signal change is instead from 50 to 100. This figure is adapted from Olsman and Goentoro 2016 [20].

small-molecule binding regulates RNA secondary structure and protein synthesis. The modular nature of RNA allows flexible implementation of post-transcriptional regulation, and has generated much interest in developing synthetic riboswitches and RNA thermometers [19].

Logarithmic sensing from protein allostery From protein regulation, we now consider the information processing capacity of proteins themselves. In the early 1960s, shortly after winning the Nobel Prize [21], the French scientist Jacques Monod, along with Franc¸ois Jacob, Jeffries Wymann and Jean-Pierre Changeux, developed an influential theoretical framework with which certain classes of protein regulation can be studied. Their work focused on describing the regulation of metabolic proteins, through a mechanism they called allostery [22,23]. Allostery has since been discovered in a wide range of biological processes including enzymes, ion channels, membrane receptors, and transcription factors [24,25,26]. An allosteric protein is one with a binding site where ligand binds and regulates the protein’s biological activity, and at least one other site. This secondary site, which binds an allosteric effector, regulates the activity of the protein without directly interacting with the primary site. The model of allostery proposed by Monod, Wyman, and www.sciencedirect.com

Changeux (MWC) describes proteins with two (or more) distinct conformational states, each functionally different. The MWC model, 

1 þ KcA

N

aðc; e0 Þ ¼  N  N 1 þ KcA þ ee0 1 þ KcI ;

assumes that the allosteric effectors alters the conformational equilibrium of the protein (as seen in the diagram in Figure 3a). In this equation, a is the fraction of protein in the active state, c is the concentration of ligand, e0 is the effective free energy between the active an inactive states, N is the total number of subunits, and KA and KI are dissociation constants for the active and inactive states, respectively. For example, hemoglobin has four subunits (N = 4) that each can bind oxygen (c). Allosteric effectors, such as carbon dioxide, alters the free energy e0 of hemoglobin and alters the equilibrium between the active (high affinity for oxygen, KA) and inactive (low affinity for oxygen, KI) states. Recent work [20] shows how allosteric proteins can, in theory, act as logarithmic sensors. Stated mathematically, this work showed that, under certain parametric Current Opinion in Biotechnology 2018, 54:72–79

76 Analytical biotechnology

constraints, the architecture in Figure 3b implements the relationship da d / dt

ln dt

ðcÞ

:

The work further shows that published biophysical measurements support that known allosteric proteins operate within this constraint and tunes its activity on log scale. It is remarkable that a seemingly complex logarithmic computation is encoded simply in the structural regulation of a protein. Therefore, beyond acting as a transport molecule (e.g. hemoglobin) or enzyme (e.g. in metabolism), the widespread allosteric proteins may also act as a quantitative sensor mediating sensing on logarithmic scale. Logarithmic sensing may offer the advantage of facilitating response to signal across many orders of magnitudes. In hemoglobin, for example, given the wide range of oxygen levels and demand throughout the body, it makes sense that hemoglobin may have evolved to respond dynamically to relative, as opposed to absolute, changes in oxygen demand. Finally, when coupled to negative feedback, logarithmic sensing may mediate an emerging recurrent phenomenon in biological signaling, where cells respond to fold changes, rather than absolute change, in signal level (as seen in Figure 3c) [27–30].

A sequestration-based integrator As a final example, we move on from the problem of sensing to the problem of control. Broadly speaking, the goal of control is to get the state of a system to match a desired set point. Feedback control does this in two steps:

first by measuring the error between the actual and desired output, and second by adjusting the input of the system to reduce the error. A common strategy in biological regulation is molecular sequestration. For instance, in RNA, a common regulation is sense-antisense pairing, where a given mRNA has a complementary RNA or DNA strand that has no direct biological function, but serves to regulate its biologically active partner [32,33]. In bacterial transcriptional regulation via sigma factors, there exist antisigma factors, whose primary role appears to be as a sequestration partner [34] (Figure 4a). Recent work by Briat et al. [31] derived a fundamental connection between sequestration mechanism and a special class of feedback, integral feedback control [35]. This insight is most concisely illustrated by mathematics. First, let us consider what an integral control is. Let x = (x1, . . . , xN) be a state vector, which could represent the concentration of N species in a network. The dynamics of the network can be described as x_ ¼ f ðx; uÞ; where f(x, u) is a function of the internal state x and a control input u, each perhaps characterized by production and degradation rates of different species in the network. It does not matter much what the details of f(x, u) are, just that the closed-loop system is stable. Our goal is to set the concentration of the species in x to a particular level xf. The task of feedback control is to find a u(x) such lim t!1 xðtÞ ¼ xf . Thus, we could define that the error e(t) = xf  x(t), and e(t) ! 0.

Figure 4

(a)

(b) Output Reference

Signal

200

Reference Signal

150 100 50

Sequestration

0

50

100

Time Current Opinion in Biotechnology

Integration through molecular sequestration. (a) A cartoon of the sequestration feedback network from [31]. Here we see a sigma factor (denoted s) which that responds to a reference signal by inducing the production of an input to the network x1. The output of the network xN then induces the production of an anti-sigma factor (denoted s) that subsequently sequesters s. (b) Reference tracking behavior in the sequestration feedback network. Here we see how the output (blue line) tracks changes in the reference signal (dashed line). The two match exactly at steady state because of integral control. Current Opinion in Biotechnology 2018, 54:72–79

www.sciencedirect.com

There’s (still) plenty of room at the bottom Olsman and Goentoro 77

The question then becomes how we pick u. An effective control strategy is to set u as the integral of the error function: Z

t

uðtÞ ¼

0

0

Z

t

eðt Þ dt ¼

0

ðxf  xðt 0 ÞÞ dt 0 :

What’s next?

0

This is known as integral feedback [35]. Integral feedback always adjusts the system to the desired state xf, since at steady state, if the system is stable, then the following must hold: d uðtÞ ¼ eðtÞ ¼ 0 dt ) xðtÞ ¼ xf : Briat et al. derived that such an integral feedback control can be implemented through molecular sequestration that is common in cellular processes [31] (Figure 4a). Let us introduce two control molecules u1 and u2, where the full system has the dynamics x_ ¼ f ðx; uÞ

ð1Þ

u_ 1 ¼ m  hu1 u2

ð2Þ

u_ 2 ¼ uxN  hu1 u2 :

ð3Þ

Here m can be taken as some external reference (i.e. an inducer of u1), u as the rate of production of u2, and h as the rate of sequestration of u1 and u2. Our goal is to set the steady-state level of xN / m, independent of any parameters that are internal to f(x, u). The sequestration reaction implies that u1 and u2 are removed from the system at the rate hu1u2. Because sequestration affects both molecules equally (and in this simple model, irreversibly), the rate of sequestration for both control molecules must be exactly equal. It follows that (see more detailed derivation in [31]), uðtÞ ¼ u1 ðtÞ  u2 ðtÞ ¼ u

Z t  m  xN ðt 0 Þ dt 0 : u 0

If we define the desired reference point xf ¼ mu and the error function e(t) = xf  xN(t), we find u_ ¼ 0 ) xN ¼

m ; u

implementing integral feedback control. We can think of the sequestration system in Eqn 1 as being composed of an actuator (u1) that affects a circuit and a sensor (u2) that www.sciencedirect.com

measures the output (xN). Thus, a seemingly simple molecular sequestration allows for a complex feedback control.

If Feynman were still alive today, we imagine that he would support our endeavor to keep pushing the ability to engineer synthetic molecular and cellular networks comparable to what is found in nature. Modern molecular biology has made rapid progress toward accomplishing his initial goal of interrogating the precise structure and mechanics of cellular machineries. The time is now ripe for bioengineers and synthetic biologists to apply this vast array of knowledge, and start building. To do so, we argue that it is important not only to push our understanding of what molecules do together, but also to appreciate diverse molecular machinery as complex computational devices. Just as a sentence is made up of words and grammar, biological circuits are made up of molecules and their associated interactions. We might view the circuits shaped by evolution as our Rosetta Stone, and current progress in engineering synthetic circuits as our first sentences spoken in this new language. In this analogy, to be able to express more complex ideas, we will not only need to develop a more sophisticated grammar (e.g. molecular interactions), but also importantly, a richer vocabulary (e.g. molecular functions). The examples we highlight here demonstrate that there is still a treasure trove from which we can learn about the vocabulary of biological circuits. Beyond the work discussed here, there has been exploration of the computational capacity of scaffold proteins acting as signaling hubs in synthetic networks [42] and of the combinatorial capacity of receptors to multiplex different signaling sources [43]. Table 1 shows a variety of other molecular mechanism/function pairs from both synthetic and natural contexts. What is becoming increasingly apparent is that biology is full of fascinating quantitative behavior encoded in relatively simple molecular mechanisms. For example, it was recently shown by Amodeo et al. [39] that the timing of the mid-blastula transition in Xenopus laevis embryos is implemented through a clever histone titration mechanism. Early on in development, transcription is heavily repressed by a large number of histones. As cells divide the total number of histones stays constant, so the number of histones per cell is halved at each division. At a threshold concentration, there are few enough histones that transcription initiates and cells rapidly undergo differentiation and reorganization. This timing is incredibly consistent, occurring approximately 7 hours post fertilization in a synchronized manner across all cells in the embryo. In this setting, we can think of the histone titration mechanism serving the function of a developmental timer. Current Opinion in Biotechnology 2018, 54:72–79

78 Analytical biotechnology

Table 1 Additional molecular mechanism/function pairs. Here we list a few other interesting molecular mechanisms that give rise to sophisticated quantitative behavior. These arise from both synthetic and natural contexts, and serve as a good reminder that biology is full of cleverly engineered behavior Molecular mechanism Inducible caspase-9 Scaffold/anti-scaffold binding Dynamic instability of microtubules Histone titration Actin-like cytoskeleton localization RNA editing

Function

Context

Source

Synthetic apoptosis Biomolecular concentration tracker Exploratory behavior Developmental timing Cell wall curvature sensing Regulation of protein kinetics

Synthetic Synthetic Natural Natural Natural Natural

Gargett et al. [36] Hsiao et al. [37] Tanaka et al. [38] Amodeo et al. [39] Ursell et al. [40] Liscovitch-Brauer et al. [41]

60 years ago, when Feynman gave his lecture, we were just beginning to understand the human capacity to manipulate and interact with digital information. Today, digital computing is so routine that we barely even notice it in our everyday life. Perhaps 60 years from now, we will have the same relationship with biological programs that we have with digital ones.

10. Blossey R, Schiessel H: Kinetic proofreading in chromatin remodeling: the case of ISWI/ACF. Biophys J 2011, 101:L30L32.

Acknowledgements

12. Wong HR: Endogenous cytoprotective mechanisms. Neuroimmune Biol 2005, 5:49-65.

We would like to thank Chris Frick and Tal Einav for insightful feedback and suggestions about the manuscript. This work was supported by the Benjamin M. Rosen Fellowship (NO), James S. McDonnell Scholar Award in Complex Systems 220020365 (to LG), and NSF CAREER Award NSF.145863 (to LG).

13. Guisbert E, Yura T, Rhodius VA, Gross CA: Convergence of molecular, modeling, and systems approaches for an understanding of the Escherichia coli heat shock response. Microbiol Mol Biol Rev 2008, 72:545-554.

References and recommended reading Papers of particular interest, published within the period of review, have been highlighted as:  of special interest  of outstanding interest 1. Feynman RP: There’s plenty of room at the bottom. Eng Sci  1960, 23:22-36. This lecture challenged scientists of the time to push the limits of what is possible at the molecular level. Feynman’s vision set a course for the future of nanotechnology and molecular biology, one that largely came to fruition in the decades to come. 2.

Alon U: Network motifs: theory and experimental approaches. Nat Rev Genet 2007, 8:450-461.

3.

Bray D: Molecular networks: the top-down view. Science 2003, 301:1864.

4.

Boccaletti S, Latora V, Moreno Y, Chavez M, Hwang D-U: Complex networks: structure and dynamics. Phys Rep 2006, 424:175-308.

5.

Newman ME: The structure and function of complex networks. SIAM Rev 2003, 45:167-256.

6.

Hopfield JJ: Kinetic proofreading: a new mechanism for reducing errors in biosynthetic processes requiring high specificity. Proc Natl Acad Sci U S A 1974, 71:4135-4139.

7.

Gue´ron M: Enhanced selectivity of enzymes by kinetic proofreading: the mystery of the cell’s accuracy in translating genes to protein has been unlocked by the discovery of a resourceful method of heightening selectivity. Am Sci 1978, 66:202-208.

8. 9.

Mckeithan TW: Kinetic proofreading in T-cell receptor signal transduction. Proc Natl Acad Sci U S A 1995, 92:5042-5046. Narlikar GJ: A proposal for kinetic proof reading by ISWI family chromatin remodeling motors. Curr Opin Chem Biol 2010, 14:660-665.

Current Opinion in Biotechnology 2018, 54:72–79

11. Rao R, Peliti L: Thermodynamics of accuracy in kinetic proofreading: dissipation and efficiency trade-offs. J Stat  Mech: Theory Exp 2015, 2015:P06001. This paper takes Hopfield’s original concept of kinetic proofreading and places it into a more general theoretical context. The authors both expand kinetic proofreading to a larger class of reaction networks, and explore new regimes of behavior where both accuracy and efficiency can be improved by proofreading.

14. El-Samad H, Kurata H, Doyle J, Gross C, Khammash M: Surviving heat shock: control strategies for robustness and performance. Proc Natl Acad Sci U S A 2005, 102:2736-2741. 15. Kurata H, El-Samad H, Iwasaki R, Ohtake H, Doyle JC, Grigorova I, Gross CA, Khammash M: Module-based analysis of robustness tradeoffs in the heat shock response system. PLoS Comput Biol 2006, 2:e59. 16. Grossman AD, Straus DB, Walter WA, Gross CA: Sigma 32 synthesis can regulate the synthesis of heat shock proteins in Escherichia coli. Genes Dev 1987, 1:179-184. 17. Morita MT, Tanaka Y, Kodama TS, Kyogoku Y, Yanagi H, Yura T: Translational induction of heat shock transcription factor §32: evidence for a built-in RNA thermosensor. Genes Dev 1999, 13:655-665. 18. Chowdhury S, Maris C, Allain FH-T, Narberhaus F: Molecular basis for temperature sensing by an RNA thermometer. EMBO J 2006, 25:2487-2497. 19. Roßmanith J, Narberhaus F: Exploring the modular nature of  riboswitches and RNA thermometers. Nucleic Acids Res 2016, 44:5410-5423. This paper demonstrates the impact of a deep molecular understanding on synthetic systems. The authors manage to combine two separate translational regulation mechanisms in a single synthetic RNA. As our ability to engineer synthetic biological components improves, this type of sophisticated molecular engineering will likely become increasingly common. 20. Olsman N, Goentoro L: Allosteric proteins as logarithmic  sensors. Proc Natl Acad Sci U S A 2016:201601791. This paper studies the connection between allosteric proteins and logarithmic sensing. The authors found that allosteric proteins are naturally well suited to logarithmic sensing, and indeed play a key role in many pathways where logarithmic response has been observed. 21. Jacob F, Monod J: Genetic regulatory mechanisms in the synthesis of proteins. J Mol Biol 1961, 3:318-356. 22. Monod J, Changeux J-P, Jacob F: Allosteric proteins and cellular control systems. J Mol Biol 1963, 6:306-329. 23. Monod J, Wyman J, Changeux J-P: On the nature of allosteric transitions: a plausible model. J Mol Biol 1965, 12:88-118. www.sciencedirect.com

There’s (still) plenty of room at the bottom Olsman and Goentoro 79

24. Changeux J-P, Edelstein SJ: Allosteric mechanisms of signal transduction. Science 2005, 308:1424-1428.

33. Morris KV, Mattick JS: The rise of regulatory RNA. Nat Rev Genet 2014, 15:423.

25. Changeux J-P, Christopoulos A: Allosteric modulation as a  unifying mechanism for receptor function and regulation. Cell 2016, 166:1084-1102. This study extends the notion of allostery to encompass a wide range of regulatory behavior. Changeaux, one of the discoverers of allostery in the 1960s, views allostery not just as a niche behavior among specialized enzymes but rather as a general property of biological molecules whose activity is strongly linked to its three-dimensional conformation.

34. Paget MS: Bacterial sigma factors and anti-sigma factors: structure, function and distribution. Biomolecules 2015, 5: 1245-1265.

26. Gunasekaran K, Ma B, Nussinov R: Is allostery an intrinsic property of all dynamic proteins? Proteins: Struct Funct Bioinform 2004, 57:433-443. 27. Goentoro L, Shoval O, Kirschner MW, Alon U: The incoherent feedforward loop can provide fold-change detection in gene regulation. Mol Cell 2009, 36:894-899. 28. Shoval O, Goentoro L, Hart Y, Mayo A, Sontag E, Alon U: Foldchange detection and scalar symmetry of sensory input fields. Proc Natl Acad Sci U S A 2010, 107:15995-16000. 29. Tu Y, Shimizu TS, Berg HC: Modeling the chemotactic response of Escherichia coli to time-varying stimuli. Proc Natl Acad Sci U S A 2008, 105:14855-14860.

35. Astro¨m KJ, Murray RM: Feedback Systems: An Introduction for Scientists and Engineers. Princeton University Press; 2010. 36. Gargett T, Brown MP: The inducible caspase-9 suicide gene system as a “safety switch” to limit on-target, off-tumor toxicities of chimeric antigen receptor T cells. Front Pharmacol 2014, 5. 37. Hsiao V, de los Santos EL, Whitaker WR, Dueber JE, Murray RM: Design and implementation of a biomolecular concentration tracker. ACS Synth Biol 2014, 4:150-161. 38. Tanaka E, Ho T, Kirschner MW: The role of microtubule dynamics in growth cone motility and axonal growth. J Cell Biol 1995, 128:139-155. 39. Amodeo AA, Jukam D, Straight AF, Skotheim JM: Histone titration against the genome sets the DNA-to-cytoplasm threshold for the Xenopus midblastula transition. Proc Natl Acad Sci U S A 2015, 112:E1086-E1095.

30. Lazova MD, Ahmed T, Bellomo D, Stocker R, Shimizu TS: Response rescaling in bacterial chemotaxis. Proc Natl Acad Sci U S A 2011, 108:13870-13875.

40. Ursell TS, Nguyen J, Monds RD, Colavin A, Billings G, Ouzounov N, Gitai Z, Shaevitz JW, Huang KC: Rod-like bacterial shape is maintained by feedback between cell curvature and cytoskeletal localization. Proc Natl Acad Sci U S A 2014, 111: E1025-E1034.

31. Briat C, Gupta A, Khammash M: Antithetic integral feedback  ensures robust perfect adaptation in noisy biomolecular networks. Cell Syst 2016, 2:15-26. This paper shows that molecular sequestration is capable of implementing integral feedback. Further, it provides theoretical insight into the nature of precise adaptation in a stochastic setting.

41. Liscovitch-Brauer N, Alon S, Porath HT, Elstein B, Unger R, Ziv T, Admon A, Levanon EY, Rosenthal JJ, Eisenberg E: Trade-off between transcriptome plasticity and genome evolution in cephalopods. Cell 2017, 169:191-202.

32. Villegas VE, Rahman MF-U, Fernandez-Barrena MG, Diao Y, Liapi E, Sonkoly E, Sta˚hle M, Pivarcsi A, Annaratone L, Sapino A et al.: Identification of novel non-coding RNA-based negative feedback regulating the expression of the oncogenic transcription factor gli1. Mol Oncol 2014, 8:912-926.

www.sciencedirect.com

42. Gordley RM, Bugaj LJ, Lim WA: Modular engineering of cellular signaling proteins and networks. Curr Opin Struct Biol 2016, 39:106-114. 43. Antebi YE, Nandagopal N, Elowitz MB: An operational view of intercellular signaling pathways. Curr Opin Syst Biol 2017, 1: 16-24.

Current Opinion in Biotechnology 2018, 54:72–79